Description

Abstract: Skin cancer remains a critical global health concern, underscoring the need for early and
accurate detection to improve patient outcomes. Advances in deep learning, particularly Convolutional Neural Networks (CNNs), have demonstrated significant potential in skin cancer diagnosis. However, challenges such as class imbalances and the trade-off between model accuracy and computational efficiency persist. This study presents a comprehensive analysis of CNN architectures for skin cancer detection using the HAM10000 dataset, comprising 10,015 dermoscopic images of seven pigmented lesions. We address class imbalances and enhance model robustness by implementing data augmentation strategies and standard preprocessing techniques, including resizing and normalization. Six state-of-the-art CNN models—VGG16, VGG19, ResNet50, MobileNet, MobileNetV2, and MobileNetV3—are systematically evaluated. Our findings reveal that ResNet50 achieves the highest accuracy and F1 score, making it a reliable choice for precise diagnosis. Conversely, MobileNetV3 excels in computational efficiency, making it well-suited for real-time applications and deployment in resource-constrained environments. This study provides critical insights into the trade-offs between diagnostic accuracy and efficiency, offering a practical framework for selecting optimal CNN models tailored to specific clinical and operational needs.

Keywords: Skin Cancer Detection, Convolutional Neural Networks, Deep Learning, VGG16, ResNet50, MobileNet, MobileNetV2, MobileNetV3.